RWDS dataset is now available online. Check out the links below and let us know if you have any feedback.

📣 Attending @cvprconference.bsky.social #CVPR2025?
Come see our work presented by @ferdaofli.com : Benchmarking Object Detectors under Real-World Distribution Shifts in Satellite Imagery.
🗓 Happening now!
📍 ExHall D, Poster #280
🔗 Project Page | 📂 RWDS Dataset
#DomainGeneralisation #CVPR2025

The #CVPR2025 main conference open access proceedings is up:

openaccess.thecvf.com/CVPR2025

Workshop papers will be posted shortly. Stay tuned...

8/ Find out more about our work here:
🌐 𝗣𝗿𝗼𝗷𝗲𝗰𝘁 𝗣𝗮𝗴𝗲: rwgai.qcri.org/rwds/
📄 𝗖𝗩𝗣𝗥 𝗣𝗮𝗽𝗲𝗿 𝗣𝗮𝗴𝗲: cvpr.thecvf.com/virtual/2025...
💾 𝗖𝗼𝗱𝗲 + 𝗗𝗮𝘁𝗮𝘀𝗲𝘁𝘀: github.com/RWGAI/RWDS
📑 𝗣𝗿𝗲𝗽𝗿𝗶𝗻𝘁: arxiv.org/abs/2503.19202
#CVPR2025

7/ 🎯 The work will also be featured in the 𝗗𝗼𝗺𝗮𝗶𝗻 𝗚𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻: 𝗘𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻, 𝗕𝗿𝗲𝗮𝗸𝘁𝗵𝗿𝗼𝘂𝗴𝗵𝘀, 𝗮𝗻𝗱 𝗙𝘂𝘁𝘂𝗿𝗲 𝗛𝗼𝗿𝗶𝘇𝗼𝗻𝘀 𝗪𝗼𝗿𝗸𝘀𝗵𝗼𝗽 𝗶𝗻 𝗖𝗩𝗣𝗥 𝟮𝟬𝟮𝟱:
📆 𝗝𝘂𝗻𝗲 𝟭𝟮𝘁𝗵, 𝟮𝟬𝟮𝟱
🕧 𝟭𝟮:𝟯𝟬 𝗽𝗺 – 𝟱:𝟬𝟬 𝗽𝗺
📍 𝗡𝗮𝘀𝗵𝘃𝗶𝗹𝗹𝗲, 𝗧𝗡 (𝗖𝗩𝗣𝗥 𝟮𝟬𝟮𝟱)
@cvprconference.bsky.social
#CVPR2025

6/📍 𝗝𝗼𝗶𝗻 𝘂𝘀 𝗶𝗻 𝗡𝗮𝘀𝗵𝘃𝗶𝗹𝗹𝗲!
𝗪𝗲'𝗹𝗹 𝗯𝗲 𝗽𝗿𝗲𝘀𝗲𝗻𝘁𝗶𝗻𝗴:
🖼️ 𝗣𝗼𝘀𝘁𝗲𝗿 #𝟮𝟴𝟬, 𝗦𝗲𝘀𝘀𝗶𝗼𝗻 𝟮
📅 𝗘𝘅𝗛𝗮𝗹𝗹 𝗗 - @cvprconference.bsky.social 𝟮𝟬𝟮𝟱, 𝗝𝘂𝗻𝗲 𝟮𝟬𝟮𝟱
#CVPR2025

5/ 🧠 Researchers: RWDS offer is a rigorous new and realistic benchmark for studying domain generalisation in object detection.
🏭 Industry: Use it to test model robustness in critical, high-stakes scenarios like climate and disaster response.

4/ We study SOTA object detectors under single- and multi-source domain generalisation. The results? 𝗜𝗻𝘁𝗼𝗹𝗲𝗿𝗮𝗯𝗹𝗲 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗱𝗿𝗼𝗽𝘀 𝗶𝗻 𝗻𝗲𝘄, 𝘂𝗻𝘀𝗲𝗲𝗻 𝗮𝗻𝗱 𝗼𝘂𝘁 𝗼𝗳 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 𝘀𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀, revealing the limitations of current AI models and the pressing need for robust, generalisable detection systems.

3/ 💡𝗥𝗪𝗗𝗦: 𝗥𝗲𝗮𝗹-𝗪𝗼𝗿𝗹𝗱 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 𝗦𝗵𝗶𝗳𝘁𝘀
We're introducing 𝗥𝗪𝗗𝗦, a benchmark suite designed to evaluate 𝗼𝘂𝘁-𝗼𝗳-𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 (𝗢𝗢𝗗) 𝗴𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘀𝗮𝘁𝗶𝗼𝗻 and 𝗼𝗽𝗲𝗻-𝘄𝗼𝗿𝗹𝗱 𝗿𝗼𝗯𝘂𝘀𝘁𝗻𝗲𝘀𝘀 in object detection, using satellite imagery from critical domains such as:
🌦️ 𝗖𝗹𝗶𝗺𝗮𝘁𝗲 𝗭𝗼𝗻𝗲𝘀
🌊 𝗙𝗹𝗼𝗼𝗱𝗲𝗱 𝗥𝗲𝗴𝗶𝗼𝗻𝘀
🌀 𝗛𝘂𝗿𝗿𝗶𝗰𝗮𝗻𝗲 𝗘𝘃𝗲𝗻𝘁𝘀

2/ This work is part of my PhD research, and I'm incredibly grateful for the mentorship and support of my PhD advisors @ferdaofli.com and Dr. David Yang 🙏

1/ 🚀 𝗡𝗲𝘄 𝗣𝘂𝗯𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗔𝗹𝗲𝗿𝘁: 𝗕𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝗶𝗻𝗴 𝗢𝗯𝗷𝗲𝗰𝘁 𝗗𝗲𝘁𝗲𝗰𝘁𝗼𝗿𝘀 𝘂𝗻𝗱𝗲𝗿 𝗥𝗲𝗮𝗹-𝗪𝗼𝗿𝗹𝗱 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 𝗦𝗵𝗶𝗳𝘁𝘀 𝗶𝗻 𝗦𝗮𝘁𝗲𝗹𝗹𝗶𝘁𝗲 𝗜𝗺𝗮𝗴𝗲𝗿𝘆! 🌍🛰️
We’re excited to announce that our paper "𝗕𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝗶𝗻𝗴 𝗢𝗯𝗷𝗲𝗰𝘁 𝗗𝗲𝘁𝗲𝗰𝘁𝗼𝗿𝘀 𝘂𝗻𝗱𝗲𝗿 𝗥𝗲𝗮𝗹-𝗪𝗼𝗿𝗹𝗱 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 𝗦𝗵𝗶𝗳𝘁𝘀 𝗶𝗻 𝗦𝗮𝘁𝗲𝗹𝗹𝗶𝘁𝗲 𝗜𝗺𝗮𝗴𝗲𝗿𝘆" has been 𝗮𝗰𝗰𝗲𝗽𝘁𝗲𝗱 𝗮𝘁 #CVPR2025! 🎉
A 🧵

4/ We study SOTA object detectors under single- and multi-source domain generalisation. The results? 𝗜𝗻𝘁𝗼𝗹𝗲𝗿𝗮𝗯𝗹𝗲 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗱𝗿𝗼𝗽𝘀 𝗶𝗻 𝗻𝗲𝘄, 𝘂𝗻𝘀𝗲𝗲𝗻 𝗮𝗻𝗱 𝗼𝘂𝘁 𝗼𝗳 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 𝘀𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀, revealing the limitations of current AI models and the pressing need for robust, generalisable detection systems.

3/ 💡𝗥𝗪𝗗𝗦: 𝗥𝗲𝗮𝗹-𝗪𝗼𝗿𝗹𝗱 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 𝗦𝗵𝗶𝗳𝘁𝘀
We're introducing 𝗥𝗪𝗗𝗦, a benchmark suite designed to evaluate 𝗼𝘂𝘁-𝗼𝗳-𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 (𝗢𝗢𝗗) 𝗴𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘀𝗮𝘁𝗶𝗼𝗻 and 𝗼𝗽𝗲𝗻-𝘄𝗼𝗿𝗹𝗱 𝗿𝗼𝗯𝘂𝘀𝘁𝗻𝗲𝘀𝘀 in object detection, using satellite imagery from critical domains such as:
🌦️ 𝗖𝗹𝗶𝗺𝗮𝘁𝗲 𝗭𝗼𝗻𝗲𝘀
🌊 𝗙𝗹𝗼𝗼𝗱𝗲𝗱 𝗥𝗲𝗴𝗶𝗼𝗻𝘀
🌀 𝗛𝘂𝗿𝗿𝗶𝗰𝗮𝗻𝗲 𝗘𝘃𝗲𝗻𝘁𝘀

2/ This work is part of my PhD research, and I'm incredibly grateful for the mentorship and support of my PhD advisors @ferdaofli.com and Dr. David Yang 🙏 #CVPR2025

Hi @cvprconference.bsky.social, when is the deadline for uploading the poster/slides/videos to the virtual website? Thanks

Thank you for organising such an insightful #ML4RS workshop at @iclr-conf.bsky.social! Great talks and discussions throughout. Looking forward to the next one 😊

Sara Al-Emadi, Yin Yang, Ferda Ofli
Benchmarking Object Detectors under Real-World Distribution Shifts in Satellite Imagery
https://arxiv.org/abs/2503.19202